Skip to content

Route Low-Confidence OCR Results to Manual Review #984

Description

@Cedarich

Complexity: Medium (150)

Description
app/ai-service/services/ocr.py and ocr_job.py return extracted text, but a low-confidence
extraction is returned exactly like a high-confidence one. The backend consequently treats
unreliable text as authoritative, and there is no signal that a human should look at the document.

Acceptance Criteria

  • OCR results carry a confidence value and a banding derived from configurable thresholds
  • Results below the threshold are flagged for review in the response
  • The flag is included in the callback payload to the backend
  • Thresholds are configurable per document type where available
  • Tests cover above-threshold, below-threshold, and missing-confidence cases

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

Stellar WaveIssues in the Stellar wave programai-servicePython FastAPI microservice for AI verification, OCR, and privacy scrubbing.evidenceWave 6 label: evidencefeatureWave 6 label: featureintegrationworkflowWorkflow automation

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions